Analysis of Forest Cover Loss in Wilpattu Forest Complex: A Remotely Sensed Change Detection and Fragmentation Analysis

: Wilpattu Forest complex is the largest protected area in Sri Lanka and it is a designated Ramsar wetland cluster. It contains a range of diverse terrestrial habitats including various forest types and a coastline with large coastal sand dunes. However, this vital ecosystem is under threat of fragmentation and degradation due to increasing human interventions. Habitat fragmentation is a serious threat to the rich biodiversity in this ecosystem. The objective of this study was to assess the loss of forest cover within the region, to examine the spatial pattern of forest cover fragmentation and to predict the potential forest cover change in 2025 using remote sensing and GIS techniques. Two Landsat 5 TM satellite images of 1992 and Landsat 8 OLI-TIRS image of 2018 were classified using unsupervised and supervised techniques to obtain three land use/ land cover classes namely, forest, non-forest and water. Forest cover change during 1992 and 2018 was analysed using classified images. Finally, future changes were modelled to assess possible threat to the forest cover. The analysis revealed that Wilpattu forest complex has lost 19,524 ha of its forest reserves to other land uses within the last 26 years and the highest impact was seen in the upper Wilpattu forest where about 7.48% area has lost from forest to other land uses. Study also proved that the total forest area is becoming more fragmented affecting the balance of the valuable natural eco-system of the area. Lack of availability in proper forest boundaries and classification issues were identified as main limitations in this study. However, it must be emphasized that immediate actions are needed to prevent further degradation of this sensitive ecosystem.


INTRODUCTION
Natural ecosystems are vital components of human and animal lives since they provide important goods and services for their existence.However, the ecosystems are under constant threat of degradation or destruction due to increasing human interactions.Forests are one of the major elements of the environment which provide the foundation for life on earth through ecological functions.It regulates the climate and water resources while providing habitats for plants and animals.According to Food and Agriculture Organization of the United Nations (FAO, 2001) the total forest cover in Sri Lanka in 1992 was about 32.2 percent of its total land area.By 2010 Sri Lanka's forest cover has reduced to 26.6% indicating a land cover of 1.7 million hectares (FAO, 2010).Ratnayake et al. (2002) concluded that four districts, namely Mannar, Puttlam, Trincomalee and Vavuniya are having serious forest cover reduction.FAO (2010) presented the deforestation rate in Sri Lanka as 1.2% during 1990-2000 in which 1.47% reduction has occurred during 2000-2005 and 0.77% reduction has occurred during 2005-2010. Hewage and Mallika (2008) stated the most serious consequences of deforestation as loss of biodiversity, irregular water supply, shortened lifespan of irrigation channels and reservoirs, soil erosion and loss of soil fertility.Forest fragmentation is one of the serious consequences of deforestation which leads to habitat fragmentation (Laurance, 1998).Fragmentation analysis can be used to quantify the forest cover change.Reddy et al. (2012) stated that fragmentation occurs when large continuous forests are divided into smaller blocks by physical barriers such as roads, agriculture, urbanization or other development activity.Three landscape level metrics can be used for the spatial characterization of the land use/land cover maps.The landscape contagion index (CONTAG) measures the degree of clumping of attributes on raster maps.The index is computed from the frequencies by which different pairs of attributes occur as adjacent pixels on a map (Ritters et al., 1996).Landscape shape index (LSI) is widely used in landscape ecology.It has been introduced to indicate the divergence of the shape of a landscape patch from an ideal shape (Circle) (Patton, 1975).Plant associations of circular shape (shortest perimeter compared to its area patches) within a certain size category are considered as the most stable and resistant in ecological sense.Shannon's diversity index (SHDI) accounts for both abundance and evenness of the species present.Estoque et al. (2016) and Subasinghe et al. (2016) cited that an increase in CONTAG indicates an increase in the degree of clumpiness of patches in terms of configuration and composition of landscape, while a decrease in CONTAG indicates an increase in the level of fragmentation.Contagion Index considers all patch types present in an image, including any present in the landscape border, if present like adjacencies.Increase in LSI value indicates an increase in the degree of landscape disaggregation or dispersion.LSI provides a standardized measure of total edge that adjusts for the size of the landscape.It measures the perimeter to area ratio for the landscape as a whole.The value of SHDI represents the amount of information per patch.SHDI is equal to 0 when the landscape has no diversity.SHDI increases as the number of different patch types increases.To obtain forecasted land use pattern of the area, multi-layer perception neural network (MLP NN) and CA-Markov modules available in the TerrSet GIS software can be used.Markov Chain process, determines the amount of change according to multi temporal land use/ land cover maps.The characteristics of different indices are given in Table 1.This study was conducted with the main objective of assessing the forest cover change in Wilpattu National Park area during the time period of 1992-2018 using remote sensing and GIS techniques.The specific objectives were to identify the spatial pattern of forest fragmentation and to predict the potential forest cover change in the future.

Study area
The Wilpattu forest belongs to North Central, North Western and Northern provinces of Sri Lanka.Wilpattu Forest Complex is a land area of 213,755 ha and is the largest remaining forested land in Sri Lanka (Senewiratne, 2017).The mean annual temperature is 27.2 °C and total annual precipitation is approximately 1000 mm.The main feature in this area is the concentration of 'villus' or 'lakes' within the Dry Monsoon forest area creating a wetland complex.Wilpattu area nourishes Kala Oya and Malwathu Oya.This forest complex is made up of many separate forest reserves, namely Marichchikaddi-Karadikkuli, Veppal, Mavillu, Vilatthaikulam (here referred as Upper Wilpattu), Wilpattu National Park, Wilpattu North Sanctuary (here referred as Wilpattu) and Thabbova Sanctuary, Veerakkulichol Sanctuary and Eluwankulam Proposed Forest Reserve (here referred as Lower Wilpattu), which was used as the study area.Wilpattu forest complex is a destination for migratory birds.Therefore, the habitat change with time would have a significant impact on biodiversity.Considering this importance, the Department of Wildlife Conservation in Sri Lanka has declared Wilpattu as a Ramsar site in 2013.Illegal settlements and encroachments are affecting the natural forest cover in this area.Environmental Foundation Limited (2017) reported that clearing of forest cover is seen widely on land which belongs to Wilpattu forest complex.Thus, there is an essential need of assessing the forest cover change in the Wilpattu forest complex in order to protect the valuable natural eco system for the future generation.

Satellite imagery and other data
Landsat satellite images with minimum cloud disturbances (<10%) and with a time interval of 26 years were obtained from the United States Geological Survey (USGS) website.Table 2 presents the characteristics of satellite images used in the study.Two satellite images of 1992 were mosaiced and produced one quality image for further processing whereas the second time point needed only one satellite image to cover the study area.Then using the forest boundary, study area was extracted.The satellite image of 1992 was classified into 45 classes using unsupervised classification technique in order to separate the pixel values into details.These classes were combined to

(A ) (B )
identify three land use/ land cover classes namely forest, non-forest and water.This 1992 image was classified using unsupervised classification technique because of the lack of ground truth data and a reliable base map to trace the points during this time point.With the help of Google Earth satellite images to collect ground truth, 2018 satellite image was classified into same three land use/ land cover classes using supervised classification.Classification accuracy assessment was carried out considering Google Earth images as a reference.
Final classified images contained three classes namely, Forest, Non Forest and Water.These classified images of two time points were subjected to further analysis to assess land use/ land cover changes, forest fragmentation, zonal analysis and land use/ land cover predictions.

Forest fragmentation analysis
The landscape-level metrics namely, CONTAG, LSI and SHDI were calculated using Fragstats 4.1 software.Figure 3 presents the flow diagram of the methodology adopted in the study.Since there were some seasonal effects on the satellite images used in extracting land use/ land cover classes, some pixels were incorrectly classified and those pixels were omitted in change analysis.About 4% of the area has changed from other land uses to forest since some adjacent areas were also named as protected area by the law.

Forest cover loss
The loss of forest cover from 1992 to 2018 is shown in Figure 5.As marked in the Figure 5, area number 1 and 2 shows the deforested areas for construction of houses.These zones are referred as area for resettlements though they are located within the forest complex.Area 3 marked in Figure 5 is the area which were already encroached and it is evident that the encroachment is further approaching towards the heart of the forest.According to the classification, all together an area of 19,524 ha forest cover loss has happened within the Wilpattu forest complex over the 26 year period.

Zonal analysis
Zonal analysis was performed by dividing the Wilpattu forest complex into three segments based on the Wilpattu national park and its upper and lower adjacent forest reserves (Figure 7).After giving all the inputs as mentioned in the above, the model simulated the future land use/ land cover with two outputs as soft prediction and hard prediction as shown in Figure 10.Soft prediction depicts the vulnerability of each pixel to transition to different land cover classes during the time period specified.In the transitional potential map, areas in red have a high potential to transition whereas areas in green have a low potential to transition.The map shows that a considerable extent of land in the forest complex has a high potential to transition indicating the vulnerability of this ecosystem for changes.In the hard prediction map, the lower Vilpattu area consists of a considerable extent of other land uses indicating forest cover loss.According to the prediction, the changes to other land uses from forest is prominent along the road network.Hence, construction of new roads within this ecosystem will make it more vulnerable to forest cover loss.

CONCLUSIONS
This study analysed the loss of forest cover within Wilpattu forest complex, examined the spatial pattern of forest cover fragmentation and predicted the potential forest cover change by 2025 using remote sensing data and GIS techniques.The analysis revealed that out of 211,259.16ha land area of Wilpattu forest complex 19,524 ha has been deforested within the past 26 years' time.Study also proved that the total forest area is more fragmented than previous increasing the risk of distorting the balance of the valuable natural eco-system of the area.
Figure 2 shows the false colour composites (FCC) (R = NIR band, G = Red band, B = Green band) of multi temporal images.

Figure 3 .
Figure 3. Flow diagram showing of the study methodology.

Figure 4 .
Figure 4. Land use/ land cover changes from 1992 to 2018 in Wilpattu Forest Complex.

Figure 5 .
Figure 5. Forest cover loss in Wilpattu Forest Complex.Fragmentation of Wilpattu forest complex Figure 6 presents the LSI, SHDI and CONTAG indices calculated for 1992 and 2018 land use/ land cover condition in Wilpattu forest complex.

Figure 7 .
Figure 7. Zonal Distribution of forest reserves in Wilpattu forest complex.

Figure 8
Figure 8 illustrates the percentages of forest and non-forest areas in each zone.From those values, frequency of each class was derived.Maximum change of the forest cover has occurred in the Upper Wilpattu area.Even the area covered by Upper Wilpattu area is smaller in extent and the highest percentage of forest cover loss is visible there.Illegal clearing of forest cover has been reported in Upper Wilpattu area.

Figure 8 .
Figure 8. Forest and other land uses (non forest and water) changes over time.

Figure
Figure 9. Driver Variables A: Distance to Roads B: Distance to Land plots C: Distance to Non Forest area

Table 3 . Land use/ land cover change matrix.
As shown in theTable 3 an area about 158451 ha of forest has remained as forest over the time.Approximately 19524 ha of land has changed from Forest to other land use/ cover classes.